Consumer Literacy and Creditworthiness

Transcription

1 Consumer Literacy and Creditworthiness Marsha Courchane Welch Consulting & Peter Zorn Freddie Mac Draft: Please do not distribute or quote without the authors permission

2 In this research we use the data gathered through a survey conducted by Freddie Mac to examine the relationship between consumer financial knowledge, consumer behavior, and credit outcomes. The survey was designed to identify factors that might contribute to any credit or payment difficulties on the part of consumers. The survey gathered detailed information on individuals self-reported use of credit, and we link that information with the individual s publicly recorded credit record. The information gathered in the survey included experience, attitudes, behaviors, and perceptions of credit; sources of information about financial matters; measures of knowledge and familiarity with finances; and demographic factors and psychological characteristics of the respondent. We use the survey responses combined with credit bureau information obtained from Experian to help understand how financial knowledge can impact a consumer s behavior and how that behavior, in turn, can lead to particular credit outcomes. 1. Initial Findings The model proposed links consumer literacy, defined as self-assessed financial knowledge or objective knowledge, to behavioral patterns such as budgeting, saving, and shopping responsibly, and to actual credit market outcomes based on information on individual credit bureau reports. This model is the one upon which much public policy implicitly relies (e.g., credit counseling, education, and early intervention). Our findings broadly support that policy structure and paradigm. We find that the single largest predictor of responsible behaviors is financial knowledge. Given this survey, self-assessed knowledge matters much more than does objective knowledge. We interpret this to mean that consumer literacy (knowledge and internalization of financial rules of thumb--don t spend more than you earn, basic budgeting, and paying bills on time) matter more than might the ability to understand a discount rate or the intricacies of credit reports. Consumer behavioral patterns are the second most predictive component of the equation predicting whether or not consumers will have impaired credit. Interestingly, while consumer literacy interventions directly impact behavior in a significant way, these do not have a direct effect on credit outcomes when examined in a reducedform framework except through formal education. Learning through courses, seminars, or from parents may affect the self-assessed perception of knowledge (and, hence, behavior, and credit outcomes), but those do not directly significantly affect outcomes. Credit counseling, on the other hand, improves consumer literacy and has a significant impact, for those with counseling, on credit outcomes, with those having previous counseling more than five years ago likely to do better. Those with recent counseling do less well, but that might reflect endogeneity the counseling results from bad credit outcomes. We find that homeownership exerts a positive influence on credit outcomes in the reduced-form equations, possibly through the impact of increased financial responsibility on creditworthiness. As observed in previous studies, African Americans and Hispanics are more likely to have impaired credit relative to whites, while Asians are likely to have less impaired credit. The largest contributors to self-assessed knowledge (in terms of marginal impact) were the group of learning variables (learn from bad times, education, and learn from school). The second group of variables with the most impact on self-assessed knowledge was the financial well-being group (income, net worth, homeownership, and income relative to parents). Other 2

3 variables had significant but small impacts. We observe slightly different results in the objective knowledge equation. While learning from bad times and formal education still have relatively large impacts, income and income relative to parents increase in the magnitude of their impacts. Counseling remains significant only if received more than five years earlier (or never). We find in the second stage that interacted knowledge (objective interacted with self-assessed knowledge) is by far the largest explanatory variable in terms of impact on behavior. The group of psychological variables included measures for whether the respondent felt as though he/she took risks, was optimistic, worried about money, worried about short-run problems, felt in control, and could cope well with stress. As a group, these variables explained a large percentage of the variance in behavior, while income factors (including safety net) and knowledge also explained significant percentages of behavior (self-control). The demographic effects (age, kids, and gender) except race were insignificant, and while race explained a small amount of the variance in behavior, so did counting on God to solve problems. We provide both structural equations and reduced-form results for credit outcomes. The largest explanatory variable in the structural, logit estimation of impaired credit is race, followed closely by behavior/self control. The presence of bad financial shocks (medical, tax, theft problems) also matters significantly. The most important of the financial variables was the presence of a financial safety net (helping weather shocks). Formal education remains important, with more schooling improving credit outcomes. One group of variables with some interest is that related to spousal interactions. When the respondent managed finances, credit outcomes were better than they were otherwise and were also better if the respondent and spouse agreed on financial matters. Similarly, spousal credit behavior impacted credit outcomes as did being left with bills by an ex-spouse. Finally, in the reduced-form equation for credit outcomes, the primary factors explaining variance in credit outcomes are race, bad events (medical, tax, and theft problems), previous bad times, counseling, formal education, and the presence of a safety net. Clearly, if issues arose, and were dealt with (earlier counseling, the safety net), then the impact of bad events and bad times on credit could be ameliorated. If these avenues (counseling, education, safety net) are less available to minorities, they might find it more difficult to prevent unfortunate events from adversely impacting credit. 2. Overview of Survey and Data Development The analysis of this paper is based on information about 12,140 respondents compiled from three sources: survey data collected from individuals who responded to a 12-page consumer credit survey (CCS) questionnaire; demographic data kept on file by Market Facts, Inc. and The NPD Group to support their panels used for surveys; and individual credit data from Experian, a consumer credit repository agency. The CCS was completed by panel members aged 20 to 40 with household incomes under $75,000. These cutoffs were chosen to represent a segment of the population for which homeownership and credit issues are important. The CCS data development process of 1999 included two firms that maintain national databases of mail survey panels: Market Facts, Inc., the lead 3

4 firm on this project, and the NPD Group. The sample frame consisted of lists of prerecruited survey respondents provided by the two survey panel companies. Since the members of these panels had agreed to complete surveys, the panels represent known populations with relatively high response rates. Both of these panels included more than 500,000 households covering the U.S. Panel information previously collected by Market Facts and NPD provided some background for sample selection but contained no credit information. Experian, Inc., a consumer credit repository, provided the credit data key to our study. The sample was selected on the basis of payment history in order to obtain an adequate sample with impaired credit. Freddie Mac assisted with survey design and initial analysis of some of the key data elements. Table 1 in Appendix A illustrates the roles of each agency in the development of the survey. Market Facts and NPD prepared files containing the names and address of all available African-American, Hispanic, and Asian panel members meeting the age and income criteria, plus a geographically balanced sample of whites. 1 In their databases, married couple mail panel households designated one head of household as the primary contact, most often the female head. Since marital status was known, survey mailings were targeted to the male or female head of married panel members to ensure more gender balance in the responses. 2 Table 2 in Appendix A provides the racial/ethnic distribution in the file sent to Experian. The file included 68,854 single household members. For the purpose of this survey, respondents are classified into only one racial/ethnic group. While Hispanics can be of any race, we categorized those with minority racial status as that race, rather than Hispanic. Next, Experian appended credit files to the name and address file provided. The process was designed to ensure confidentiality of consumer credit information in that neither the survey panel companies nor Freddie Mac could match an individual s credit record to that individual s name, address, or other unique identifying information. Experian matched and provided credit information for 85,597 individual householders and spouses (excluding the out-of-range Asians) or 91,223 (including out-of-range Asians). These matched files were used for selection of the survey samples. Table 3 (in Appendix A) shows the race/ethnic distribution of the panel members and their spouses after matching. 3 If first name and race/ethnicity of spouses of panel members were not available, the following decision rules were adopted for the credit reporting agency to use for the purpose of achieving the greatest accuracy in the appending of credit records. An adult of the opposite 1 For the Asian sample, there were insufficient persons to achieve the sampling objectives. However, national estimates from the survey exclude Asians who are outside the age or income specifications. 2 The CCS uses individuals, not households, as the unit of observation. Many, if not most, credit decisions involve more than one family or household member, and the survey includes several questions about the role of the spouse/partner in the questionnaire. While records can be merged, each individual consumer has his or her own payment and credit record. Attitudes, perceptions, and opinions, moreover, are inherently personal and require that survey participants answer as an individual, not a household. 3 In mail panel households, one head of household is designated as the person to be the first recipient of all survey mailings/contacts. Typically, it is the female in married couple households. It was necessary to include panel member spouses as designated respondents to minimize gender bias. 4

5 gender living at the same address as a married panel member was assumed to be the spouse (or partner) and was deemed eligible for the sample. An individual classified as the spouse of the panel member was assumed to share the race/ethnicity category of the respective spouse/partner when this information was absent for the spouse. If no spouse was found for the married panelist, it is shown as spouse not found in Table 3. After credit records for all panel members and married spouses (when found) were appended to the files, the files were forwarded to Freddie Mac, where each panelist was grouped into a credit quality group. The sampling plan partitioned by race/ethnicity (white, African- American, Hispanic, and Asians) and payment history (impaired, indeterminate, good, and nonmatches). 4 Each panelist in the sample was categorized in one of three credit quality groups ( buckets ) using actual payment behavior extracted from credit files. The definitions of credit buckets used for our analysis are as follows: Impaired: Good: Respondents are impaired if they meet the following conditions: 90 days or more late or in derogatory status on one trade line in the past 24 months 30 days or more late on another trade line in the past 24 months To be impaired, this group fell behind on their payments on at least two separate accounts in the last two years, and in at least one of these instances, they became as late as 90 days delinquent. People are put into the good bucket if they meet all of the following conditions: days late on no more than one trade line in the past 24 months Never 90 days or more late or in derogatory status in the past 24 months Fewer than four instances of ever being 30 days late on a trade line No public records ever filed on any trade line Indeterminate: People are put into the indeterminate credit bucket if they fit none of the other buckets. Non-matches: People are put into the nonmatch bucket if Experian did not return credit record variables. 5 People are excluded from our analysis if they have neither an impaired nor good payment record (i.e., have an indeterminate payment record) or if Experian was unable to match them to a credit record. Observations are weighted to adjust for this bucket definition. Table 4 (Appendix A) shows the distribution, by data development phase, of race/ethnicity for panel members and married spouses for which Experian was able to match their records. 4 The goal was to obtain approximately 1,000 surveys in each of the 12 cells of this four-by-four matrix. 5 Nonmatches could include those individuals not covered by Experian but may have records in other credit repository agencies or those individuals covered by Experian but whose name didn t match. In the original sampling design, the nonmatches were combined with the good. 5

6 Market Facts and NPD mailed out almost 23,000 surveys to the pre-identified panel members and spouses. Slightly higher proportions were mailed to African-American and Hispanic panelists, as well as to male spouses of female panel members, in order to compensate for the known tendency of these groups to have lower survey response rates than whites and females, respectively. All of the Asian members on the two panels were sent surveys, since this is a segment that is dramatically underrepresented in the panels. Table 5 in Appendix A provides the distributions by race/ethnicity and payment history of the sample to whom surveys were mailed. With respect to race/ethnicity, the respondent was classified based on the reported race/ethnicity from the survey. If the respondent did not answer the race/ethnicity questions, race/ethnicity was based on the panel company information or assumed for spouses to be the same as respondents. As previously stated, for the purpose of this survey, respondents are classified into only one racial/ethnic group. Thus, Hispanics who indicate that they are white are classified as Hispanic. The overwhelming majority of Hispanics fall in this racial group. Hispanics who report that they are African-American are grouped as African-American. Finally, Hispanics who report that they are Asian are classified as Hispanic. Table 6 (in Appendix A) shows the result of the original race/ethnicity by self-reported race/ethnicity. A total of 12,140 questionnaires were returned, providing a response rate of more than 52 percent. The distribution of the final sample is provided in Table 7 in the Appendix with responses categorized by race/ethnicity, marital status, and credit bucket. The final sample allowed for a two-year relaxation on the age criteria. Respondents reporting their age on the questionnaire as between 18 and 42 were included. In addition, as income often is highly variable and reporting of income often is inaccurate, survey respondents were not excluded from the final data file based exclusively on the income they reported on the survey. The raw data were then weighted to approximate the national composition of the U.S. population of householders aged 20 to 40 with household incomes of less than $75,000. This allows for analysis of the data both by sample cell and at the aggregate level. The data have been weighted by those factors on which the sample was originally stratified, as well as some additional characteristics that tend to be highly correlated with issues of financial well being. These include: Race/ethnicity by gender Race ethnicity by payment history ( bucket ) Age Household income Household size A form of marginal-based weighting known as rim weighting was used to simultaneously adjust for each of these factors. The weighting logic is based on a procedure of multivariate post hoc balancing, in which the population target distributions of the weighting factors are inputs to the weighting program. The weighting program uses an optimizing algorithm that seeks the best set of weights for all combinations of variable values or cells. When applied to the data, this yields the univariate target distributions on each variable as closely as possible. (See Appendix B for the target population distributions used.) 6

7 After Market Facts cleaned, edited, and weighted the data, files were sent to Freddie Mac where alternative credit bucket definitions were developed and for which Market Facts provided new weights. For this analysis we use the definition of impaired for which the individual must be days delinquent in at least one trade line, and 90 or more days delinquent or derogatory status in another trade line in the past 24 months. Observations included in the estimation are weighted to reflect the U.S. population after accounting for the choice-based sampling employed when administering the survey. Those values are presented in Table 8. Table 8. Final Sample New Definition of Impaired Phase VII. Final Sample by Race/Ethnicity and Credit Bucket 4 RACE/ETHNICITY Final Sample Pct of Race Credit Bucket N Pct of Final Sample African American 3, Impaired 1, Indeterminate 1,040 9 Good Non-matches Asian* 1, Impaired Indeterminate Good Non-matches Hispanic 2, Impaired Indeterminate Good Non-matches White 4, Impaired 1,051 9 Indeterminate 1, Good 1, Non-matches Totals 12, Impaired 3, Indeterminate 3, Good 3, Non-matches 1,

8 3. Analysis We estimate a recursive model with credit outcomes, a function of financial behavior that, in turn, is a function of self-assessed or objective financial knowledge. We will estimate each of the equations in the recursive model with all respondents and then by racial or ethnic category. We estimate both structural models, including actual values of knowledge in the behavior equation and both knowledge and behavior in the outcome equation, and reduced-form models without the inclusion of the other dependent variables as predictors. We use OLS procedures (with a Type III option) for the knowledge and behavior equations and an ordered logit (with a Type III option) for the outcome equations. The Type III option provides a statistical test that allows us to obtain F-test results (for OLS) or likelihood ratio test results (Wald statistics for logit) for the class of variables in addition to the individual parameter estimates for the values of the class that are typically obtained without that option. First, we estimate self-assessed and objective knowledge in separate equations as functions of predictors of that knowledge. These predictors include reference variables (education, age, gender, family status, race), income and wealth-related variables (income, wealth, employment status, homeownership status, income relative to family of origin, and expected or past variation in income), and other survey variables that might contribute to the attainment of knowledge. These include learning from parents, regular savings, after school jobs, student loans, credit counseling, presence of a safety net, and money management courses. Next, we estimate behaviors as a function of financial knowledge and additional factors that will affect behavior. The dependent variable is a summary of financial self-control composed from answers to several survey questions. We include whether the respondent follows a budget or saves or invests money from each paycheck. We also include controlling spending, paying bills on time, planning for financial future, providing for self and family, only buying things that are necessary, and borrowing for things that are not important. We assign negative value to buying things that are not affordable. Not included in the financial knowledge equations but included here are psychological or physical health factors that might influence financial control behaviors. We include self-descriptive variables, such as risk-taking, optimism, the influence of religion, and gambling. We also include stress-related impacts that might affect behavior. These include variables such as the amount of worry about money, spending, and the respondent s financial situation, smoking, and stress symptoms (nightmares, migraines, insomnia, stomach or back pain, extreme tiredness or fatigue, or feelings of inadequacy). Other explanatory variables include feelings of control or lack thereof (inadequacy, helplessness, or lack of control) relative to feeling in control (I can do anything I set my mind to and what happens depends on me) should impact behavior. We call these variables locus of control. Here we also include spousal credit behaviors. Our financial equation estimates the financial outcome for each respondent. The dependent variable is impaired credit using the alternative definition presented above. For this equation, we estimate financial outcome using an ordered logit. In addition to the inclusion of financial self-control and predictors discussed above, we include in the financial outcome equation variables describing spousal impacts (from current or ex-spouse), divorce and the status of unpaid bills following divorce, and intra-household management and financial decisionmaking. We also include factors grouped to indicate learning from the school of hard knocks. These survey questions elicit responses on bad financial events (eviction, NSF checks, utility 8

9 cancellations, credit denials, creditor calls, repossession, late payments, collections, and bankruptcy). Finally, we include bad external events, such as major medical expenses, theft or property destruction, or major legal or tax problems. Model of Creditworthiness Source of Knowledge Parental Influence Income & Variability Home Ownership Employment Credit Counseling After School Job Education/Loan Gender/Age/Children Marital Status Race Financial Knowledge Objective Subjective Financial Behavior (Self Control) Budgeting and Saving Ability to Control Finances Additional Variables: Self Assessed Knowledge Objective Knowledge Attitudes Locus of Control Ability to Cope Financial Safety Net Feelings Income/Net Worth Ex-Spouse Effects Credit Outcomes Negative Events Financial Stress and Strain Additional Variables: Behavior Bad Events School of Hard Knocks Divorce Impacts Management of Finances Agreement about Finances 9

10 Overall, we find that the equations fit the model well. The key explanatory variable for behavior is knowledge and that for credit outcomes is behavior. In this respect, the vision that motivates intervention for credit counseling and educational programs will have value. To the extent that more learning affects knowledge and that knowledge impacts how persons behave with respect to saving, budgeting, and spending, we expect better behavior to result in better credit outcomes (lower impaired credit). Individual equation results have some interesting features, but the main finding is that there are important policy reasons to continue advocating improved financial awareness. Turning to individual equations, we first present results for the knowledge equations in Tables 9 and 10. Self-assessed (SE) knowledge is based on the respondents beliefs about what they know, while objective knowledge is based on answers to particular financial questions. Generally speaking, there seems to be support for the view that one can enhance financial knowledge by providing learning opportunities through a variety of sources. We find a relatively large impact on self-assessed knowledge coming from learning experiences, including learn from bad times (school of hard knocks). Interestingly, this is more important in terms of explaining the variance in SE knowledge than the existence of previous bad credit outcomes, including repossessions, bankruptcies, and creditor calls. In addition to this, learning from formal education has a significant impact (the more education, the higher the impact), as does taking financial training courses in school (learn from school) and financial seminars (learn from seminars). Credit counseling can also impact knowledge. Above and beyond the impact on behavior, knowledge does not have a direct impact on credit outcomes. As shown in the reducedform results in Table 15, the set of learning variables, beyond formal education, do not individually significantly explain outcomes. Income and wealth also affect own estimations of knowledge, with higher incomes and higher net worth associated with more knowledge. Homeownership exerts a positive influence on SE knowledge, as does having higher income levels compared to one s parents. Variance in income has some effect, with having a decrease in income in the past two years the largest effect from the variance measures. The presence of a financial safety net matters considerably, with being likely able to cover bills for three months having the largest impact on SE knowledge. This may reflect a tendency toward saving and budgeting reflected in the dependent variable. Finally, we see a large effect from the two credit card variable associated with usage and payment patterns. For previous credit card usage, those obtaining a card at the youngest age learned, marginally, the most. This may not reflect positive credit outcomes, simply an increase in financial knowledge that comes from use of credit. Card payment patterns considerably impact knowledge, with those paying only the minimum, and having small balances, not learning much. Those who pay in full (up to balances of $5,000) learn a significant amount from the experience. Once we look at the credit outcomes variable, having a credit card does not have a significant impact but credit payment patterns remain significant. 10

11 Table 9 Estimations for Self-Assessed Knowledge Parameter Estimate T-Statistic Pr > t Intercept <.0001 Card Use <.001 under < or older <.0001 never had card Card Pay <.001 min, >=$5000 due min, $ min, <$1000 due > min, >=$5000 due <.0001 > min, $ > min, <$1000 due >> min, >=$ <.0001 >> min, $ <.0001 >> min, <$1000 due <.0001 in full, >=$ in full, $ in full, <$ do not use cards Bad Credit Outcomes <.001 no problems <.0001 problems Student Loan no student loan student loan Learn from Bad Time <.001 times little <.0001 times some <.0001 times a lot Counseling <.001 no, never yes, <2 years yes, 3-5 years yes, >5 years After School Job <.001 seldom <.0001 sometimes often Income <.001 under $25, <.0001 $25,000 to $44, $45,000 or more 11

12 Net Worth <.001 under $10, <.0001 $10,000 to $49, $50,000 or more Own or Rent <.001 own <.0001 rent <.001 Y Compared to Parent worse off same better off do not know Safety Net <.001 unlikely <.0001 neutral <.0001 likely Employment Status <.001 self-employed <.0001 part-time not working student retired/disabled full-time Y Variance seldom sometimes often Net Y Chg in Past 2 Yrs decreased <.0001 stayed the same <.0001 increased Expected Net Y Chg unlikely neutral likely Ex-Spouse Bills <.001 no ex-spouse left bills left no bills Education <.001 some school <.0001 high school <.0001 some college associate degree finished college 12

13 Learn from School <.001 a little <.0001 some <.0001 a lot Learn from Seminars <.001 a little <.0001 some a lot Gender male female Age < 30 years old >= 30 years old Kids no kids kids Race <.001 Hispanic African-American Asian White Results from the objective knowledge equation are similar in terms of significance of particular explanatory variables, with credit card use and credit card payment patterns, income and learning (from bad times or from formal education) still important. We see a smaller impact from counseling and no impact from homeownership as compared to SE knowledge. Later, in looking at the behavioral equation, Table 11, we observe that knowledge has a strong impact on behavior (with objective knowledge and self-assessed knowledge interacted). Table 10 Estimations for Objective Knowledge Parameter Estimate T-Statistic Pr > t Intercept <.0001 Card Use under or older never had a card 13

14 Card Pay <.0001 min, >=$5000 due min, $ min, <$1000 due > min, >=$5000 due <.0001 > min, $ <.0001 > min, <$1000 due >> min, >=$ <.0001 >> min, $ <.0001 >> min, <$1000 due in full, >=$ in full, $ <.0001 in full, <$ <.0001 do not use cards Bad Credit Outcomes no problems problems Student Loan no student loan student loan Learn from Bad Times <.0001 little <.0001 some a lot Counseling no, never yes, <2 years yes, 3-5 years yes, >5 years After School Job seldom sometimes often Income <.0001 under $25, <.0001 $25,000 to $44, <.0001 $45,000 or more Net Worth under $10, $10,000 to $49, $50,000 or more Own or Rent own rent 14

15 Y Compared to Parent <.0001 worse off <.0001 same <.0001 better off <.0001 do not know Safety Net <.0001 unlikely neutral <.0001 likely Employment Status self employed part-time not working student retired/disabled full-time Y Variance seldom sometimes often Net Y Chg in Past 2 Yrs decreased stayed the same increased Expected Net Y Chg unlikely neutral likely Ex-Spouse Bills no ex-spouse left bills left no bills Education <.0001 some school <.0001 high school <.0001 some college <.0001 associate degree finished college Learn from School little some a lot Learn from Seminars little some a lot 15

16 Gender <.0001 male <.0001 female Age < 30 years old >= 30 years old Kids no kids kids Race <.0001 Hispanic African-American <.0001 Asian White After modeling the consumer literacy equations, we turn to the behavior equation. The dependent variable here is an index of types of good behavior, such as saving regularly, budgeting, controlling spending, and paying bills on time. Included as explanatory variables are psychological factors, income-related factors, consumer literacy factors, and demographics. By far the most important determinant of behavior/self control was knowledge (self-assessed interacted with objective). The respondent had less self-control when the knowledge was either stated as little or some in either self-assessed or objective categories, compared to having a fair amount of knowledge in both categories. The set of psychological factors also had an expected large impact on financial behavior. A respondent behaves better if more optimistic, taking fewer risks, not worrying too much about money, and being able to cope well. Feeling in control (locus of control) has a significant but relatively small effect. The income-related measures do matter, but with income relative to parents and the existence of a safety net being considerably more important than actual income, net worth, or homeownership. Here, we see some direct effects from formal education and a small effect from learning about financial matters at school, but the learning variables are less important here (likely because of their influence on knowledge). 16

17 Table 11 Estimations for Behavior/Self Control Parameter Estimate T value Pr > t Intercept <.0001 Knowledge Interacted <.0001 both very little <.0001 very little, some <.0001 very little, fair amount some, very little both some <.0001 some, fair amount <.0001 fair, very little fair amount, some both fair amount Take Risks <.0001 slightly <.0001 somewhat <.0001 well Optimistic <.0001 slightly <.0001 somewhat <.0001 well Count on God <.0001 slightly <.0001 somewhat well Worry in SR slightly somewhat well Go to Church seldom sometimes often Gamble seldom sometimes often Locus of Control internal neutral external 17

18 Stress seldom sometimes often Smoke do not smoke smoke Worry about Money <.0001 very little <.0001 some <.0001 a fair amount Cope <.0001 well <.0001 OK <.0001 poorly Student Loan no student loans student loan Income under $25, $25,000 to $44, $45,000 or more Net Worth <.0001 under $10, <.0001 $10,000 to $49, $50,000 or more Own or Rent own rent Y Compared to Parents <.0001 worse off same better off <.0001 do not know Safety Net <.0001 unlikely <.0001 neutral <.0001 likely Employment Status self employed part-time not working student retired/disabled full-time 18

19 Income Variance seldom sometimes often Net Y Chg in Past 2 Yrs <.0001 decreased <.0001 stayed the same <.0001 increased Expected Net Income Chg unlikely neutral likely Spouse <.0001 no spouse <.0001 fair <.0001 okay <.0001 good Ex-Spouse no ex-spouse fair okay good Ex-Spouse Bills left bills left no bills Education <.0001 some school high school <.0001 some college <.0001 associate degree finished college Learn from School a little some a lot Learn from Seminars a little some a lot Gender male female Age < 30 years old >= 30 years old 19

20 Kids no kids kids Race Hispanic African-American Asian White Finally, we examine the impacts of behavior on impaired credit. As we noted earlier, behavior is the second most important variable in explaining credit outcomes (with only race more important). Here, as the impacts of learning are already measured by the literacy impact on knowledge and behavior, these variables do not have further direct effects. We do see for the first time some significant impacts from demographic variables (age, kids, gender) all with the expected signs. Race is the single most important factor in explaining impaired credit outcomes, somewhat surprising not in its impact but in its magnitude. We continue to observe the importance of relative income compared to the previous generation and the safety net (which may be provided by the previous generation). Also of note are the effects from those variables relating to income and employment uncertainty (unemployment or income fall, net income change in past two years, or expected net income change), all of which have significant impacts on credit outcomes. Marital accord leads to better outcomes than does marital discord (agree on finances and good spousal behavior). Table 12 Structural Estimations of Credit Outcomes Parameter Estimate ChiSq Pr > ChiSq Intercept Behavior/Self Control <.0001 poor <.0001 okay <.0001 good very good Events <.0001 no <.0001 one of them two or three of them Ex-Spouse Bills yes, unpaid bill nes, no bills no Unemployment or Y Fall no <.0001 one of them both of them 20

21 Student Loan <.0001 no student loan <.0001 student loan Income under $25, $25,000 to $44, $45,000 or more Net Worth under $10, $10,000 to $49, $50,000 or more Own or Rent <.0001 own <.0001 rent <.0001 Y Compared to Parents worse off same better off do not know Safety Net <.0001 unlikely <.0001 neutral <.0001 likely Employment Status self employed part-time not working student retired/disabled full-time Income Variance seldom sometimes often Net Y Chg in Past 2 Yrs <.0001 decreased <.0001 stayed the same increased Expected Net Income Chg unlikely neutral likely Spouse <.0001 no spouse <.0001 fair okay <.0001 good 21

22 Ex-Spouse fair okay good Manage Finances mostly spouse equally mostly me Agree on Finances seldom sometimes often Education <.0001 some school <.0001 high school <.0001 some college <.0001 associate degree <.0001 finished college Learn from School little some a lot Learn from Seminars little some a lot Gender male female Age < 30 years old >= 30 years old Kids <.0001 no kids <.0001 kids Race <.0001 Hispanic <.0001 African-American <.0001 Asian White Our final table of results presents the reduced-form equation for impaired credit. We wanted to compare the direct effects of the explanatory variables on impaired credit to their impacts when entered through the other structural equations, consumer literacy and behavior. Here we find that the psychological variables, by themselves, do not generally influence credit outcomes directly. As expected, bad financial shocks (bad events) do continue to matter as do 22

23 previous bad times (learn from bad times) and previous bad credit outcomes (bad credit outcomes). While talking to parents may not matter, observing their behavior, saving as a child, and getting an after school job all do matter in small but significant ways. This argues for early financial training (e.g., adding money management to health classes in high school). The income and net worth variables are reduced to insignificance in this form, as are the interventionist measures (learning from school or seminars). A surprise is the strong impact of recent credit counseling on impaired credit outcomes, although this likely follows from the need to get counseling due to already extant credit issues. Table 13 Reduced Form Estimations of Credit Outcomes Estimate Chi Sq Pr>ChiSq Intercept Card Use under or older never had a card Card Pay <.0001 min, >=$5000 due min, $ min, <$1000 due > min, >=$5000 due <.0001 > min, $ <.0001 > min, <$1000 due >> min, >=$ <.0001 >> min, $ <.0001 >> min, <$1000 due in full, >=$ in full, $ <.0001 in full, <$ <.0001 do not use cards Bad Events <.0001 no <.0001 one of them two or three of them Ex-Spouse Bills yes, unpaid bill yes, no bills no Unemployment or Y Fall no one of them both of them 23

24 Take Risks slightly somewhat well Optimistic slightly somewhat well Count on God slightly somewhat well Worry in SR slightly somewhat well Go to Church seldom sometimes often Gamble seldom sometimes often Locus of Control internal neutral external Stress seldom sometimes often Smoke do not smoke smoke Worry about Money very little some a fair amount Cope well OK poorly Bad Credit Outcomes <.0001 no problems <.0001 problems 24

25 Student Loan no student loan student loan Learn from Bad Times <.0001 little <.0001 some <.0001 a lot Counseling <.0001 no, never yes, <2 years <.0001 yes, 3-5 years yes, >5 years Talk to Parents disagree neutral agree Parents-Good Mgmt disagree neutral agree Save as Child disagree neutral agree Learn from Parents little some a lot After School Job seldom sometimes often Income under $25, $25,000 to $44, $45,000 or more Net Worth under $10, $10,000 to $49, $50,000 or more Own or Rent <.0001 own <.0001 rent 25

26 Y Compared to Parents worse off same better off do not know Safety Net <.0001 unlikely <.0001 neutral likely Employment Status self employed part-time not working student retired/disabled full-time Variance in Income seldom sometimes often Net Y Chg in Past 2 Yrs decreased stayed the same increased Expected Net Income Chg unlikely neutral likely Spouse <.0001 no spouse <.0001 fair okay <.0001 good Ex-Spouse fair okay good Manage Finances mostly spouse equally mostly me Agree on Finances seldom sometimes often 26

27 Education <.0001 some school high school some college <.0001 associate degree finished college Learn from School little some a lot Learn from Seminars little some a lot Gender male female Age < 30 years old >= 30 years old Kids <.0001 no kids <.0001 kids Race <.0001 Hispanic <.0001 African-American <.0001 Asian White We look finally at the marginal impacts of several groups of the explanatory variables. These groups are composed of several different questions from the survey responses, grouped to reflect similar types of expected impacts. We include here many of the exogenous variables, such as bad events and income or unemployment changes, as well as previous bad credit outcomes. We combine several financial variables into an income and wealth category and combine several spousal-related variables, since those have been mentioned as important by focus groups. The groups follow below. 27

28 Table 14 Variable Groups credit bucket outcomes self_control behaviors bad event, dissolve, U or Y fall external events bad credit, student loan, hard knocks school of hard knocks income, net worth, own or rent income & wealth Y now compared to parents, safety net income & wealth employment income & wealth income variance, past net Y variability in income & wealth expected net Y variability in income & wealth spouse, manage, agree spouse and ex-spouse credit behaviors ex-spouse bills spouse and ex-spouse credit behaviors education, learn from school education variables learn from seminars education variables gender age kids demographic variables race demographic variables We examine the impact of the variable group, relative to a model that first includes only an intercept, and then as a group added to the full model that includes all other variable groups. In this way we can look at the sole impact of the set of variables in contributing to the explained variation, and then we can also look, at the margin, at whether that group of variables, given the variation already explained by the other variable groups, still has a large impact on credit outcomes. Those findings follow in Table 15. Table 15 Impacts of Groups of Variables on Credit Outcomes Percentage Impact Impact of Variable Effect when Controlling for Variable Group Group Alone Other Variables Behaviors 32.37% 5.68% External Events 32.51% 5.79% Hard Knocks 1.70% 1.31% Income and Wealth 53.06% 12.81% Variability in Income and Wealth 9.84% 2.67% Spousal Behaviors 29.06% 3.33% Education 16.75% 4.76% Demographics 25.24% 10.56% The group of variables most important in explaining the observed credit outcomes is that composed of income and wealth information (income, wealth, homeownership, income relative to parents, and employment status). This group contributes over 50 percent to the explained variation. After controlling for all other variables, we still see that the marginal impact of this group is 13 percent. Behaviors (self-control) and external events (medical, theft, tax, loss of job) 28

29 also have a sole contribution of over 30 percent in explaining credit outcomes. In the full model, their net contributions are around 6 percent each as a group. Demographics, particularly race, is not one of the most important groups when considered solely (at only 25 percent), but after controlling for the other groups, this set of variables continues to add over 10 percent to explaining credit outcomes. One issue that must be faced to resolve credit issues is to develop, on the part of a consumer, an accurate self-assessment of their credit situation. That is, improving credit profiles requires recognition by borrowers that there is a need to improve. This requires both understanding what contributes to impaired credit and understanding what, if anything, can be done to change the underlying circumstances contributing to credit outcomes. In this next table, we present information on how well respondents self-assess (question 13) compared to their actual credit performance (defined as impaired or good ). We present this information by racial group, as we observe some distinct differences in what percentages of respondents in a group self-assess correctly (or wrongly) compared to other groups. Table 16 Self Assessed Credit Compared to Actual Credit SELF-ASSESSED CREDIT HISPANIC ACTUAL CREDIT. Very Bad Bad Average Good Very Good Impaired Percent Row Percent Column Percent Good Percent Row Percent Column Percent AFRICAN AMERICAN ACTUAL CREDIT. Very Bad Bad Average Good Very Good Impaired Percent Row Percent Column Percent Good Percent Row Percent Column Percent ASIAN ACTUAL CREDIT. Very Bad Bad Average Good Very Good Impaired Percent Row Percent Column Percent Good Percent Row Percent Column Percent

30 WHITE ACTUAL CREDIT. Very Bad Bad Average Good Very Good Impaired Percent Row Percent Column Percent Good Percent Row Percent Column Percent If we compare the percentages, by race, of those who had impaired credit with those who believed they had bad or very bad credit, we see only minor differences. The African American respondents correctly identified their impaired credit 55 percent of the time, while whites correctly self-assessed impaired credit 59 percent of the time, Hispanics 60 percent, and Asians 64 percent. The African American group is nearly 10 percent more likely (55 percent compared to 59 percent) to have an inaccurate perception of their own impaired credit. In terms of good credit, we observe additional significant differences by race. African Americans correctly selfassessed (good or very good categories) only 65 percent of the time, while whites correctly selfassessed good credit 76 percent of the time, Hispanics 74 percent, and Asians 80 percent. Perhaps because the African Americans are less likely able to understand that they are less likely to seek credit from, for example, prime lenders, because they do not assess a high probability of success to their applications. Conclusions The findings in this paper improve our understanding of the complex interactions between acquisition of financial knowledge, behavior resulting from that knowledge, background, financial status, and psychological profile and credit outcomes. We find throughout this research that financial knowledge can improve financial behavior and behaviors (either current or perhaps influenced by past familial interactions) do significantly affect the probability of having impaired rather than good credit. We hope that this research helps to illuminate the need for consumer literacy and that it helps quantify the forms in which that literacy might be enhanced. Credit outcomes can change and increased knowledge (and awareness of that knowledge) can lead to better financial behavior and improved credit outcomes. 30

31 REFERENCES Cosslett, Stephen. R. "Efficient Estimation of Discrete-Chioce Models," in Structural Analysis of Discrete Data with Econometric Applications, edited by Manski and McFadden, Faber, Ronald J. and Thomas C. O Guinn. Compulsive Consumption and Credit Abuse, Journal of Consumer Policy, 11, 1988, Kennickell, Arthur B. Codebook for the 1995 Survey of Consumer Finances, Federal Reserve Board, February Kiplinger s, Personal Financial Survey, May 22, Lazarus, Richard S. And Susan Folkman. Stress, Appraisal, and Coping, Springer Publishing: New York, 1984, Lea, Stephen E.G., Paul Webley, and Catherine M. Walker. Psychological Factors in Consumer Debt: Money Management, Economic Socialization, and Credit Use, Journal of Economic Psychology, 16, 1995, Medina, Jose F., Joel Saegert, and Alicia Gresham. Comparison of Mexican-American and Anglo-American Attitudes Toward Money, Journal of Consumer Affairs, 30 (1), 1996, Rotter, J.B. Some Problems and Misconceptions Related to the Construct of Internal versus External Control of Reinforcement, Journal of Consulting and Counseling Psychology, 43, 1966, Thaler, Richard H. and George Loewenstein. Intertemporal Choice, The Winner s Curse: Paradoxes and Anomalies of Economic Life, ed. Richard H. Thaler, Princeton: Princeton University Press, Tokunaga, Howard. The Use and Abuse of Consumer Credit: Application of Psychological Theory and Research, Journal of Economic Psychology, 14, 1993,

32 Appendix A. Phase and Sample Tables. Table 1. CCS Data Development Process PHASE Market Facts & NPD Experian, Inc. Freddie Mac Phase I Obtain names and addresses of the panel members (See Table 2) within age & income categories Send address files (with spouse) to Experian Phase II Identify Spouse of panelist Append files with credit records of panel members and married spouses, when found (See Table 3) Remove names and addresses for privacy and forward the files to Freddie Mac Phase III Create credit quality buckets using credit variables from Experian (See Table 4) Remove credit records and forward the files to Market Facts, Inc Phase IV Draw samples & mail surveys (see Table 5) Collect, edit and code the data (See Table 7) Weight the data Phase V Merge the survey response files with credit bucket files Add geographic codes Develop alternative buckets Phase VI Rim weight the alternative buckets and send files to Freddie Mac Phase VII Merge the alternative buckets with the data

33 Table 2. Race & Ethnicity of File Sent to Experian PHASE I: Generate Names and Addresses of Panelist Race/Ethnicity N Pct of Pop Status N Pct of Race African American 14,937 22% Unmarried 10,375 69% Married 4,625 31% Asian* 7,456 11% Unmarried 2,997 40% Married 4,459 60% Hispanic 16,551 24% Unmarried 6,993 42% Married 9,558 58% White 29,910 43% Unmarried 11,880 40% Married 18,030 60% Total 68, % Unmarried 32,245 53% Married 36,609 47% *Includes out-of-range Asians. Excluding out-of-range, Asians total is 64,784. Race/ethnicity is based on the panel company information or assumed for spouses to be the same as respondent. 33

34 Table 3. Race/Ethnicity Distribution of Population After Experian Appended PHASE I: Generate Panelist Sample PHASE II: Append Credit and Add Spouses RACE/ETHNICITY Marital Status of Panelist N Pct by Marital Status Individual N Pct of Race Append Credit Data N Pct of individuals African American Unmarried 10,375 69% Panelist 10,375 57% Credit Matched 8,688 84% Credit Non-Matched 1,687 16% Married 4,562 31% Panelist 4,562 25% Credit Matched 3,791 83% Credit Non-Matched % Spouse Found 3,442 19% Credit Matched 2,091 61% Credit Non-Matched 1,351 39% Spouse Not Found 1,120 25% N/A N/A N/A Asian* Unmarried 2,997 40% Panelist 2,997 32% Credit Matched 2,561 86% Credit Non-Matched % Married 4,459 60% Panelist 4,459 48% Credit Matched 3,820 86% Credit Non-Matched % Spouse Found 1,831 20% Credit Matched 1,489 81% Credit Non-Matched % Spouse Not Found 2,628 59% N/A N/A N/A Hispanic Unmarried 6,993 42% Panelist 6,993 33% Credit Matched 5,809 83% Credit Non-Matched 1,184 17% Married 9,558 58% Panelist 9,558 45% Credit Matched 8,176 85% Credit Non-Matched 1,382 15% Spouse Found 4,763 22% Credit Matched 3,350 70% Credit Non-Matched 1,413 30% Spouse Not Found 4,795 50% N/A N/A N/A White Unmarried 11,880 40% Panelist 11,880 28% Credit Matched 10,033 85% Credit Non-Matched 1,847 16% Married 18,030 60% Panelist 18,030 43% Credit Matched 16,028 89% Credit Non-Matched 2,002 11% Spouse Found 12,333 29% Credit Matched 9,480 77% Credit Non-Matched 2,853 23% Spouse Not Found 5,697 32% N/A N/A N/A 34

35 Table 3 (cont.): Race/Ethnicity Distribution of Population After Experian Appended PHASE I: Generate Panelist Sample PHASE II: Append Credit and Add Spouses RACE/ETHNICITY Marital Status of Panelist N Pct by Marital Status Individual N Pct of Race Append Credit Data N Pct of individuals Totals Unmarried 32,245 47% Panelist 32,245 35% Credit Matched 27,091 84% Credit Non-Matched 5,154 16% Married 36,609 53% Panelist 36,609 40% Credit Matched 31,815 87% Total Panelists 68,854 Credit Non-Matched 4,794 13% Spouse Found 22,369 25% Credit Matched 16,410 73% Credit Non-Matched 5,959 27% Spouse Not Found 14,240 39% N/A N/A N/A Total Credit Matched 75,316 83% Total Credit Non-matched 15,907 17% Total Population 91, % *Including out-of-range Asians. Excluding out-of-range Asians the total is 85,597 35

36 Table 4. Race/Ethnicity Based on Panel Company Information by Credit Bucket PHASE I: Generate Panelist PHASE II: Append Credit and Add Spouses PHASE III: Create Credit Buckets RACE/ETHNICITY Marital Status N Pct of Individual N Pct of Appending Credit N Percent Bucket N Percent of Panelist Race Race Data African American Unmarried 10,375 69% Panelist 10,375 56% Credit Matched 8,688 84% Impaired 5,226 50% Indeterminate 1,798 17% Good 1,664 16% Credit Non-Matched 1,687 16% Non-matches 1,687 16% Married 4,562 31% Panelist 4,562 25% Credit Matched 3,791 83% Impaired 2,349 51% Indeterminate % Good % Credit Non-Matched % Non-matches % Spouse Found 3,442 19% Credit Matched 2,091 61% Impaired 1,129 33% Indeterminate % Good % Credit Non-Matched 1,351 39% Non-matches 1,351 39% Spouse Not Found 1,120 25% N/A N/A N/A N/A N/A N/A Asian* Unmarried 2,997 40% Panelist 2,997 32% Credit Matched 2,561 85% Impaired % Indeterminate % Good 1,741 58% Credit Non-Matched % Non-matches % Married 4,459 60% Panelist 4,459 48% Credit Matched 3,820 86% Impaired % Indeterminate 411 9% Good 2,674 60% Credit Non-Matched % Non-matches % Spouse Found 1,831 20% Credit Matched 1,489 81% Impaired % Indeterminate 172 9% Good 1,055 58% Credit Non-Matched % Non-matches % Spouse Not Found 2,628 59% N/A N/A N/A N/A N/A N/A *Includes all Asians. The non-match credit bucket were originally combined with the good credit bucket 36

37 RACE/ETHNICITY Table 4 (cont.): Race/Ethnicity Based on Panel Company Information by Credit Bucket PHASE I: Generate Panelist PHASE II: Append Credit and Add Spouses PHASE III: Create Credit Buckets Marital Status of Panelist N Pct of Race Individual N Pct of Race Appending Credit Data N Percent Bucket N Percent Hispanic Unmarried 6,993 42% Panelist 6,993 33% Credit Matched 5,809 83% Impaired 2,589 37% Indeterminate 1,171 17% Good 2,049 29% Credit Non-Matched 1,184 17% Non-matches 1,184 17% Married 9,558 58% Panelist 9,558 45% Credit Matched 8,176 86% Impaired 3,315 35% Indeterminate 1,405 15% Good 3,456 36% Credit Non-Matched 1,382 14% Non-matches 1,382 14% Spouse Found 4,763 22% Credit Matched 3,350 70% Impaired 1,319 28% Indeterminate % Good 1,416 30% Credit Non-Matched 1,413 30% Non-matches 1,413 30% Spouse Not Found 4,795 50% N/A N/A N/A N/A N/A N/A White Unmarried 11,880 40% Panelist 11,880 28% Credit Matched 10,033 84% Impaired 3,434 29% Indeterminate 1,775 15% Good 4,824 41% Credit Non-Matched 1,847 16% Non-matches 1,847 16% Married 18,030 60% Panelist 18,030 43% Credit Matched 16,028 89% Impaired 4,965 28% Indeterminate 2,200 12% Good 8,863 49% Credit Non-Matched 2,002 11% Non-matches 2,002 11% Spouse Found 12,333 29% Credit Matched 9,480 77% Impaired 2,832 23% Indeterminate 1,250 10% Good 5,398 44% Credit Non-Matched 2,853 23% Non-matches 2,853 23% Spouse Not Found 5,697 32% N/A N/A N/A N/A N/A N/A 37

38 RACE/ETHNICITY Table 4 (cont.): Total Race/Ethnicity Based on Panel Company Information by Credit Bucket PHASE I: Generate Panelist PHASE II: Append Credit and Add Spouses PHASE III: Create Credit Buckets Marital Status N Pct of Individual N Pct of Appending Credit N Percent Bucket N Percent of Panelist Race Race Data Totals Unmarried 32,245 47% Panelist 32,245 35% Credit Matched 27,091 84% Impaired 11,761 36% Indeterminate 5,052 16% Good 10,278 32% Credit Non-Matched 5,154 16% Non-matches 5,154 16% Married 36,609 53% Panelist 36,609 40% Credit Matched 31,815 87% Impaired 11,364 31% Indeterminate 4,721 13% Good 15,730 43% Total Panelists 68,854 Credit Non-Matched 4,794 13% Non-matches 4,794 13% Spouse Found 22,369 25% Credit Matched 16,410 73% Impaired 5,542 25% Indeterminate 2,489 11% Good 8,379 37% Credit Non-Matched 5,959 27% Non-matches 5,959 27% Spouse Not Found 14,240 39% N/A N/A N/A N/A N/A N/A Total Credit Matched 75,316 83% Impaired 28,667 38% Indeterminate 12,262 16% Good 34,387 46% Total Credit Non- 15,907 17% Non-matches 15,907 21% Matched Total Population 91, % Total Population 91, % 38

39 Table 5. Starting Samples Race/Ethnicity Based on Panel Company Information by Credit Bucket PHASE IV: Draw Sample and Mail Surveys RACE/ETHNICITY Starting Sample Pct of Race Credit Bucket N Pct of Starting Sample African American 6,591 36% Impaired 2,679 41% Indeterminate 1,572 24% Good 1,014 15% Non-matches 1,326 20% Asian* 3,712 40% Impaired % Indeterminate % Good 1,764 48% Non-matches % Hispanic 6,525 31% Impaired 2,495 38% Indeterminate 1,670 26% Good 1,498 23% Non-matches % White 5,884 14% Impaired 1,986 34% Indeterminate 1,908 32% Good 1,473 25% Non-matches 517 9% Totals 22,712 25% Impaired 8,116 36% Indeterminate 5,699 25% Good 5,749 25% Non-matches 3,148 14% *Includes only Asians meeting age and income criteria of target population. Including all Asians in the starting sample would result in Asian totals of: 1,465 (Impaired), 836 (Indeterminate), 2,345 (Good), and 607 (Nonmatches). Original Table 6. Original Race/Ethnicity by Self-Reported Race/Ethnicity Self-Defined Race/Ethnicity Race/Ethnicity White African-American Hispanic Asian TOTAL White 3, ,538 African-American 106 3, ,838 Hispanic , ,459 Asian* ,097 1,305 TOTAL 4,656 3,747 2,471 1,266 12,140 39

40 Table 7. Final Sample of Re-classified Race/Ethnicity by Credit Buckets and Marital Status RACE/ETHNICITY (N) African American (3,747) Asian* (1,266) Individual PHASE IV: Final Sample of Re-classified Race Pct of Marital N Credit Bucket N Status Pct of Starting Sample Unmarried Panelist 1,914 51% Impaired % Indeterminate % Good % Non-matches % Married Panelist 1,210 32% Impaired % Indeterminate % Good % Non-matches % Spouse % Impaired % Indeterminate 82 13% Good 92 15% Non-matches % Unmarried Panelist % Impaired 92 18% Indeterminate 60 12% Good % Non-matches 52 10% Married Panelist % Impaired % Indeterminate 64 15% Good % Non-matches 63 14% Spouse % Impaired 80 25% Indeterminate 47 15% Good % Non-matches 49 15% * Includes only Asians meeting age and income criteria of target population. Including all Asians in the final sample results in Asian totals as follows: 445 (Impaired), 258 (Indeterminate), 851 (Good), and 208(Non-matches). 40

41 Table 7. Final Sample of Re-classified Race/Ethnicity by Credit Buckets and Marital Status PHASE IV: Final Sample of Re-classified Race RACE/ETHNICITY Individual N Pct of Marital Pct of Final Credit Bucket N Status Sample Hispanic Unmarried (2,471) Panelist % Impaired % Indeterminate % Good 96 15% Non-matches 58 9% Married Panelist 1,047 45% Impaired % Indeterminate % Good % Non-matches 91 9% Spouse Found % Impaired % Indeterminate % Good % Non-matches % White (4,656) Unmarried Panelist % Impaired % Indeterminate % Good % Non-matches 62 8% Married Panelist 1,987 43% Impaired % Indeterminate % Good % Non-matches 159 8% Spouse 1,928 29% Impaired % Indeterminate % Good % Non-matches % 41

42 RACE/ETHNICITY Table 7. Final Sample Phase IV. Final Sample by Race/Ethnicity and Credit Bucket Final Sample Pct of Race Credit Bucket N Pct of Final Sample African American 3, Impaired 1, Indeterminate Good Non-matches Asian* 1, Impaired Indeterminate Good Non-matches Hispanic 2, Impaired Indeterminate Good Non-matches White 4, Impaired Indeterminate 1, Good 1, Non-matches Totals 12, Impaired 4, Indeterminate 3, Good 3, Non-matches

43 Appendix B. Weighted Variable Frequencies Table B1 Dependent Variables Category Frequency (Weighted) Percent (Weighted) Outcome impaired good Self Assessed Knowledge very little some a fair amount Objective Knowledge very little some a fair amount Self Control poor okay good very good Table B2 Explanatory Variables Category Frequency Percent Card Use under or older never had a card Card Pay minimum due more than minimum in full do not use cards Bad Credit Outcomes Student Loan no problems problems no student loan student loan

44 Learn from Bad Times Counseling After School Job Income Net Worth Own or Rent a little some a lot no, never yes, <2 years yes, 3-5 years yes, >5 years seldom sometimes often under $25, $25,000 to $44, $45,000 or more under $10, $10,000 to $49, $50,000 or more Own Rent Y Compared to Parents Safety Net Employment Status worse off same better off do not know unlikely neutral likely self employed part-time not working student retired/disabled full-time

45 Income Variance Net Y Chg in Past 2 Yrs Expected Net Income Chg Spouse Ex-Spouse Manage Finances Agree on Finances seldom sometimes often decreased stayed the same increased unlikely neutral likely no spouse fair okay good no ex-spouse fair okay good no spouse mostly spouse equally mostly self no spouse seldon sometimes often Ex-Spouse Bills no ex-spouse left bills left no bills

46 Education Learn from School Learn from Seminars Gender Age Kids Race Bad Events U or Y Fall Take Risks some school finished HS some college associate degree finished college nothing some a lot nothing some a lot Male Female < 30 years old >= 30 years old no kids kids Hispanic African-American Asian White no one of them two or three of them no one of them both of them slightly (describes me) somewhat well

47 Optimistic Count on God Worry in SR Go to Church Gamble Locus of Control Stress Smoke Worry about Money Cope Talk to Parents slightly (describes me) somewhat well slightly (describes me) somewhat well slightly (describes me) somewhat well seldom sometimes often seldom sometimes often internal neutral external seldom sometimes often do not smoke smoke very little some a fair amount well okay poor disagree neutral agree

48 Parents-Good Mgmt disagree neutral agree Save as Child disagree neutral agree